Predictive Maintenance of Permanent Magnet (PM) Wind

6 August 2014 Predictive maintenance of Permanent Magnet Wind Generator

AMES Research Group, Department of Electrical Engineering

Predictive Maintenance of Permanent

Magnet (PM) Wind Generators

by

Oladapo OGIDI

Supervisor: A/Prof. Paul Barendse

Co-Supervisor: A/Prof. Azeem Khan

Presented at:

Postgraduate Renewable Energy Symposium

17-18 July 2014, Stellenbosch University



Presentation Outline

1. Introduction

2. Maintenance of PM Wind Generators

3. Fault Diagnostic Techniques for Electrical Machines

4. PM Machine Faults Types and Analysis

5. Detection of Static Eccentricity (SE)

6. Conclusion



It is a fast growing renewable energy source

From 1997 – 2012, wind power had an average growth rate of 28%

As at 2012 year-end:

- There was 282GW global installed wind power capacity

- It accounted for 27% of total global renewable energy supply

Source: Global Wind Energy Council (GWEC)

1. Introduction

Why Wind Energy?

GWEC envisages that wind power would meet between 7.7% - 8.3% of global

electrical demand in 2020 and 14.1% - 15.8% in 2030.

It is renewable and abundant

It has now established itself as a mainstream renewable electricity generation

source, and plays a central role in an increasing number of countries’ immediate

and longer term energy plans

1



Suitable for direct-drive wind turbine system which is based on a simple

principle, fewer rotating components compared with geared-drive system:

- reduced mechanical stress

- increased technical service life

- increased power production of the turbines

- could be deployed in regions with low wind speed

PM Generators in Wind Turbines

There is an increase in manufacturing shift towards PM generators and direct-drive

system

Global cumulative installed capacity

2004

World market: 100%

Top 10 manufacturers: >95%

Geared generator system: >78%

Direct-drive (DD) generator system: <20%

DD excitation system: PM<5%

EE>15%

Introduction contd.

improved maintainability and efficiency

Optimization of wind resources

and better yield

Field excitation provided by PMs:

- excitation loss is reduced

- power electronics converters could be integrated since their price has

been declining over the years

2

2011

78.8

<75%

>25%

PM>EE (actual figure not available)



1. Reliability issues resulting from the operation of PM generators in both normal and

hash environmental conditions, e.g. demagnetization.

Challenges of Deploying and Maintaining PM Generators in Wind Turbines

3. Long downtime associated with generator component in wind turbines, table 1.

2. Difficulty in accessing wind farms in remote locations for repairs, e.g. offshore.

Components

Annual Failure

Rate

Down Time Per Failure

(Days)

Electrical System 0.57 < 2

Electronic Control 0.43 < 2

Sensors 0.25 < 2

Hydraulic System 0.23 < 2

Yaw System 0.18 2 – 3

Rotor Hub 0.17 > 4

Mechanical Brake 0.13 2 - 3

Generator 0.11 > 7

Rotor Blades 0.11 > 3

Gear Box 0.1 6 - 7

Support & Housing 0.1 > 3

Drive Train 0.05 > 5

Table 1. Components and failure indices

3

2. Maintenance of PM Wind Generators

Challenges 1-3 can be

overcome by implementing

predictive maintenance

strategies. It is based on

online condition monitoring

using electrical machine

fault diagnostic techniques.

Predictive maintenance

combines the best features

of both preventive and

corrective maintenance.



Furthermore, unlike induction machines, there are no standard online diagnostic

techniques for PM machines and for the wind industry.

Intermittent nature of wind resources causes recurring transient state in wind

generators, thus, this need to be overcome in the development of fault diagnostic

schemes for predictive maintenance purposes.

Major Challenge in Applying them in PM Wind Generators

3. Fault Diagnostic Techniques for Electrical Machines

The diagnostic technique widely used is:

Signal-based mechanical vibration analysis

acoustic analysis

shock pulse monitoring

temperature monitoring

motor current signature analysis

electromagnetic field monitoring

using search coils

oil and gas analysis, infrared analysis

partial discharge measurement

RF injection

Types of Fault Diagnostic Techniques

4



5

Only signal-based techniques using various standard signal processing methods

are employed at industry level for the following reasons:

1. Machine signals can be measured and thus severity of faults can be quantified.

2. Accommodate in-situ and non-invasive analysis.

3. Expert knowledge is not required in making decision.

Others are:

Artificial Intelligence-based Neural network

Fuzzy logic

Genetic algorithm

Machine-theory-based Winding function approach

Magnetic equivalent circuit

Simulation-based

Finite element analysis

Fault Diagnostic Techniques of Electrical Machines contd.



6

Standard Signal Processing Methods

Table 2. Periodogram (for steady-state condition)

Method Advantages Disadvantages

Periodogram

(It uses FFT; a

non-parametric

method)

a. Ease of

implementation.

a. Long measurement time is needed

(=>30s) and as such stationary signal

is not guaranteed.

b. Large number of samples (=>100k)

need to be collected, thus increasing

memory and cost requirements of

data acquisition devices.

To overcome the problem of long measurement duration and large number of

samples, the parametric method alternative is proposed. Examples are Estimation

of Signal Parameters via Rotational Invariance Technique (ESPRIT), Multiple Signal

Classification (MUSIC) etc.


‐ Both ESPRIT and MUSIC algorithms use same technique but ESPRIT is superior to

MUSIC in accuracy and as such it is preferred.

Non-Parametric (Fourier-based): Periodogram and Short-time Fourier Transform



‐ The basic principle is that if a signal depends on a finite set of parameters, then

all of its statistical properties including its power spectrum can be expressed in

terms of these parameters. This is done using parametric models that relate to

the eigenvector decomposition of the correlation matrix to estimate the discrete

part of the spectrum.

Estimation of Signal Parameters via Rotational Invariance Technique (ESPRIT)


ESPRIT

(a parametric

method)

a. Very high accuracy.

b. Very short measurement

time (=<3s) and as such

stationary signal is better

guaranteed.

c. Small number of samples is

required (=<10 k).

a. Computational time is longer

than FFT.

(ESPRIT-60s;10ks)

(FFT-1s;600ks)

b. Algorithm is more complex;

computes Eigen-value and

covariance matrix.

Table 3. ESPRIT (for steady-state condition)


7



Table 4. Short-Time Fourier Transform (for transient condition)


Short-Time

Fourier

Transform (STFT)

a. It is based on the

conventional

Fourier

Transform:

Simple and

robust

a. The averages of time and frequency are

never correctly given, hence, trade–off

between the time and frequency

localization of the spectrogram.

b. It scrambles the energy distribution of

the signal with that of the window

function.

Others are: Hilbert Transform, Wavelet Transform, Cohen-Class Distribution

(Wigner-Ville, Choi-Williams…) etc.

Improved Cohen-Class Distribution (for transient condition)

To overcome the disadvantages above, a type of Cohen-Class Distribution namely,

Zhao-Atlas Marks (ZAM) Distribution is proposed.

- It is a quadratic-time frequency technique that computes the energy level of a

signal and its distribution across time-frequency domain.


8



9


Zhao-Atlas Marks

(ZAM) Distribution

a. Independent of the

choice of window and as

such, no cross-term

artifact in time-

frequency domain.

b. Better resolution than

STFT

a. It is a spectrogram and as

such, the exact magnitudes

of energy spectrum is

difficult to obtain.


Table 5. Zhao-Atlas Marks (ZAM) Distribution (for transient condition)



Common Faults Associated with PM Wind Generators Table 6. PM generated fault types

FAULTS ELECTRICAL MECHANICAL Types Winding faults

Core or

lamination

Magnet damage:

Broken/displaced

PMs

Eccentricity

Broken/misaligned

shaft

Bearing Faults

Sub-types Short circuit

ф-ф,3-ф,ф-grd

inter – turn

Open circuit

High impedance

–

–

Static

Dynamic

–

–

Parts of

generator

affected

Stator

Stator

Rotor

Rotor Rotor Shaft

Rotor

Shaft

Rotor

Frequency of

occurrence

Frequent

Very Rare

- Frequent

(Axial flux)

Rare

(Radial flux)

Rare

Frequent

Severity (danger

and damage)

High

Cata-

strophic

Medium Low Medium Medium

10

4. PM Machine Faults Types and Analysis



rotor

airgap

stator

Fig.1. Healthy (left) Static Eccentric (right)

SEF increases as outer radius/diameter

increases, thus, AFPM machine is more

susceptible to SE than RFPM machine

because of its larger aspect ratio.

𝑔 = original air-gap length

𝑟 = 𝑔𝑚𝑎𝑥 − 𝑔 = 𝑔 − 𝑔𝑚𝑖𝑛

𝛽 = sin−1𝑟

𝑅

𝑆𝐸𝐹 =𝑅.sin 𝛽

𝑔× 100% 𝒈𝟏 = 𝒈 𝟏 ±

𝒓

𝒈

𝒈𝟏 is the effective air-gap length

In the absence of eccentricity,

𝒈𝟏= 𝒈

airgap

𝛽

Position of mechanical rotation

airgap

stator symmetrical axis

rotor symmetrical axis

𝛽 = deflection angle

𝛽

5. Detection of Static Eccentricity (SE) 2D Analysis: Axial-flux PM Machine

𝑆𝐸𝐹 =𝑟

𝑔× 100% (SEF: Static Eccentricity Factor)

11

Fig.2. Cross-section of asymmetric air gap (triangle)

𝛽

𝛽

𝑔𝑚𝑖𝑛

𝑔𝑚𝑎𝑥

𝑟

R

R

𝛽



Steady State Detection Using ESPRIT

SE is a mechanical fault, therefore vibration (mechanical signal) is chosen to be

analysed. Current and voltage monitoring are also fitting since SE results in

asymmetric airgap flux distribution.

Vibration Monitoring

Detection of Static Eccentricity (SE) contd.

SE Vibration Fault Frequencies

- The 2nd harmonic of the magnetic force. This is present in the vibration spectrum.

- Vibration harmonics (increased magnitude) arise at frequencies relating to the

following:

‘Twice-the-line’ frequency, i.e., 2𝑓, where 𝑓 is the fundamental frequency of the

line current.

Fig.3. Fault harmonics at 2𝑓 Fig.4. Fault at 350Hz; frequency of the 2nd harmonic of the magnetic force

12



Transient Detection Using ZAM

Fig.5. Healthy spectrogram Fig.6. Spectrogram for 40% SE

Fig.7. Spectrogram for 60% SE

The ellipses indicate visible harmonics at 58Hz and 350Hz; i.e., 2𝑓 and 2nd harmonic

frequency of the magnetic force respectively

Increase in spectrum

energy level across

time-frequency domain

is obtained in

conditions of SE.

Fig.8. Energy spectrum of the vibratory response

Detection of Static Eccentricity (SE) contd.

13



14

6. Conclusions

The techniques explored are effective for vibration monitoring for both steady-

state and transient conditions.

Predictive maintenance can reliably be built on condition monitoring strategies

based on the techniques.

The problem of fault detection due to non-stationary phenomenon associated

with wind generators can be overcome using ESPRIT as spectral estimator.

ZAM distribution offers high energy resolution and as such is effective in

computing the energy distribution of vibration signal across time-frequency

domain.

AFPM machine

torque transducer

prime mover

Fig.9. Test rig (left) and data acquisition devices for the signal capture (right)



Thanks for listening!

Documents

Predictive Maintenance of Permanent Magnet (PM) Wind